Architectures and algorithms for iconic-to-symbolic transformations

摘要

Most special architectures for image processing handle image-based (iconic) transformations efficiently but are relatively inefficient for transforming images into more symbolic representations such as lists of prominent lines, contour-description strings, or graphs that express the major spatial relationships in the image. Consequently some new architectures have been proposed to improve such operations. Here a representative sample of such architectures is presented in order to elucidate the relevant design issues. Algorithms for these architectures are described and compared. Some new algorithms for iconic-to-symbolic transformations on pipelined systems are presented.